Artemij Amiranashvili

Papers

Total Citations

H-Index

About

Artemij Amiranashvili is a researcher at the forefront of reinforcement learning (RL), with a particular focus on bridging the gap between simulation and real-world robotic control. His work centers on two critical challenges: enabling agents to perceive and act upon motion in dynamic environments, and improving the efficiency of training under domain randomization—a key technique for sim-to-real transfer. In his highly cited 2018 and 2019 papers (12 and 10 citations, respectively), Amiranashvili pioneered the explicit treatment of motion perception in RL, arguing that learned controllers must actively model object movement rather than relying on implicit learning. This work laid a foundation for more robust policy development in environments with moving objects. More recently, his 2021 paper (2 citations) tackled the practical hurdle of domain randomization, which, while essential for transfer, often introduces noise that degrades RL training. By proposing a pre-training strategy for deep RL agents, he offered a method to stabilize learning under visual variation, directly advancing the feasibility of deploying learned policies on real robots. Amiranashvili’s contributions are essential reading for anyone working in model-free RL, sim-to-real transfer, or robotic perception in dynamic settings.